The semiconductor fabrication process requires various processing apparatuses. For example, in a film forming or an etching processing of an object (e.g., a semiconductor wafer, a glass substrate etc.), a processing apparatus such as a plasma processing apparatus is used widely. In the plasma processing apparatus, a processing gas introduced in a sealed processing chamber is converted into a plasma to perform plasma processing on the surface of an object such as a semiconductor wafer. Thus, reaction by-products and the like produced by the plasma processing deposit on the inner wall surface of the processing chamber during the plasma processing, which is performed continuously, to affect the plasma state. As a result, the plasma state is altered slightly. Since such changes in the plasma state would affect the processing result of wafers from etching, the plasma state changes or the processing results must be monitored to maintain the quality of the processing at all times.
Thus, a test wafer is prepared in advance and an etching processing is regularly performed on the test wafer, so that the processing apparatus status can be checked based on the processing result (e.g., amount of erosion, uniformity or the like of the test wafer).
However, when the processing apparatus status is checked based on a test wafer, a large number of test wafers are needed and then the test wafers must be processed by using the processing apparatus whose processing condition is to be monitored and respective processing results must be measured. The preparation of such a large number of test wafers and the measurement of their processing results have been an issue because this practice entails additional costly labor and time.
Further, there has been proposed a method for monitoring the processing of a plasma processing apparatus, as disclosed in Japanese Patent Application Laid-open No. 10-125660. In this method, a model equation correlating a plasma processing characteristic with an electric signal reflecting the plasma state is generated by using a test wafer before a processing; and the plasma processing characteristic is predicted by substituting in the model equation a detection value of an electric signal obtained when processing an actual wafer.
However, when generating a model equation correlating electric signals with plasma processing characteristics, if the same model is used when the plasma processing is performed under other operation conditions, e.g., different etching conditions, the prediction of results or the detection of abnormality, which is obtained by using the model, becomes unreliable. For example, when a model A is generated by performing the plasma processing under an operating condition A, if the model A is applied to monitor the processing result or the abnormality detection of the plasma state when performing the plasma processing under a different operating condition B, the overall accuracy cannot be trusted. Thus, an optimum model is dependent on specific operating conditions. Consequently, it becomes necessary to generate a model equation by multivariate analysis for each operating conditions. If one model equation generated in one operating condition is applied to another operating condition, the accuracy of the prediction of results or the detection of abnormality may not be sufficiently reliable. In the semiconductor fabrication industry, the model equations must be generated to handle various operating conditions since the trend for small quantity batch production is increasingly dominating the market.
However, if the model is newly generated every time the operating condition is altered, the cost of operation processing required for the model generation would then increase. For example, there is a cost that wet cycle data needs to be collected for several weeks or data collection needs to be conducted numerous times while changing the operating condition by using an orthogonal array produced based on a center condition and then such collected data must be analyzed.
Further, in order to formulate a model to predict processing results on a semiconductor wafer or the like, it is required to obtain measurement data, e.g., sensor data, and to obtain the processing results such as shape, etching rate and the like after the semiconductor wafer needs to be unloaded. In this case, it is also necessary to obtain the sensor data and the processing results of the wafer under numerous conditions by using the orthogonal array and the like. If the model is newly formulated every time the operating conditions are changed, the burden of measurement is also significantly increased.
Still further, the above-described problems are not limited to cases involving electric signals when predicting the plasma processing characteristics, but the same problems are confronted also in case of adopting emission from the plasma; emission absorption by radicals in the plasma; analysis of exhaust gas components in a processing chamber by using quadrupole mass spectrometric method or FT-IR (Fourier Transform-Infrared spectroscopy); and measurement data of a film thickness of a polymer deposited on an inner wall of the processing chamber obtained through the use of acoustic element and the like.